Proceedings of the Annual Meeting of the Cognitive Science Society

Many methods exist to learn causal models from data, as causal relationships form the basis for successful actions. These methods are frequently evaluated based on the completeness of the models they can infer. Yet, there is a gap between the highly complete and potentially complex models algorithms can learn and the types of information people can use successfully to make decisions. To address this we conduct two experiments to understand how the size and features of causal models influence how well they can be used for decision-making. In Experiment 1 we systematically vary model size for a series of topics, finding that there is a negative and linear relationship between causal model size and decision accuracy. In Experiment 2 we examine how model structure influences decisions, varying whether the models include feedback loops, again finding that smaller models lead to better choices, and that feedback loops are also beneficial.